Sentence pair

Pick a source/translation pair

Drift

Drift amount: 0% Drift type

Auto sweep

Sweep speed: 1.0×

Live stats

Quality score—

Cosine similarity—

Alignment distance—

Tokens changed—

Score trend—

Source (EN)—

Translation (target)—

Each word is mapped to a point in a shared 20-dimensional synthetic embedding space: a source word and its correct translation are noisy copies of the same underlying "concept" vector, so a faithful translation lands close to the source. Sentence vectors are the average of word + word-pair vectors; quality score = cosine similarity between the source and translation sentence vectors, rescaled to 0–100. Drag "Drift amount" to substitute, reorder or omit words and watch the real cosine similarity respond.